The model picker is a dead end
gmays
12 points
4 comments
August 27, 2026
Related Discussions
Found 5 related stories in 57.8ms across 4,692 title embeddings via pgvector HNSW
- Choosing an AI model: one prompt, 11 models, different results toddmorey · 192 pts · August 13, 2026 · 49% similar
- Agent Is Not the Model joejag · 66 pts · August 24, 2026 · 49% similar
- Using an open model feels surprisingly good msaltz · 173 pts · July 28, 2026 · 47% similar
- No, local models will not win m0rde · 18 pts · August 11, 2026 · 44% similar
- Models Are Getting Dumber on Purpose hruvhwe · 300 pts · August 16, 2026 · 44% similar
Discussion Highlights (3 comments)
altcognito
This article certainly talks a lot about their brand and their approach. Understandable! Giving up model selection is giving up some level of control and understanding. I have no doubts over the long haul what is being said here is true, but I'm confident the tooling to perform model selection will be pushed upstream, not downstream. Model vendors have a strong incentive to produce controllable results and will follow their own benchmarks. The entire "self-improving" concept is reliant on getting better at getting to a solid result with the right efforts and subagents.
hodder
I don’t get how this differs from any other optimization harness. The problem is clear, but I’m not sure on what this company is doing that is unique to solve it.
yoggies_bro
More AI written bloat. why should I bother reading it if you can't bother making it readable? It's like reading a student's essay who had a page of information and is stretching it to fill 5 pages.